Project
active
CNN Comparison on Static Hand Gestures
A comparative study of 15 pretrained ImageNet CNNs for static sign-language gesture recognition, with interpretability and error analysis.
What it is
One reproducible notebook per architecture (DenseNet, ConvNeXt, EfficientNet, MobileNet and ResNet families, among others), each with confusion matrices, Grad-CAM attention maps, LIME explanations and t-SNE embeddings. Cross-model misclassification tables isolate which gesture pairs every architecture confuses.